IILLS

IILLS predicts virus-receptor interactions by integrating known virus-receptor interaction data and receptor amino acid sequences to identify known and hidden receptor-binding pairs for studying viral entry and therapeutic target discovery.

Key Features:

• Initial Interaction Scores: Uses initial interaction scores derived from neighboring interactions to establish candidate virus-receptor pairings.
• Laplacian Regularized Least Square Algorithm: Applies a Laplacian regularized least squares algorithm to refine predictions and control model complexity via regularization.
• Gaussian Interaction Profile (GIP) Kernel: Computes virus similarity using a Gaussian Interaction Profile kernel and computes receptor GIP similarity from interaction profiles.
• Receptor Sequence Similarity: Computes receptor amino acid sequence similarity and incorporates it alongside GIP-based similarities.
• Similarity Integration: Determines final receptor similarity based on prediction results with priority given to sequence similarity.
• Validation and Benchmarking: Evaluates predictive performance with 10-fold cross-validation (10CV) and leave-one-out cross-validation (LOOCV), reporting AUCs and comparisons to BRWH, LapRLS, and CMF.

Scientific Applications:

• Hidden interaction prediction: Predicts previously unreported virus-receptor interactions to expand known interaction networks.
• Therapeutic target identification: Aids identification of receptors that are potential antiviral therapeutic targets.
• Mechanistic investigation of viral entry: Supports studies of viral receptor-binding mechanisms and host specificity.
• Antiviral strategy development: Informs development and prioritization of antiviral interventions by highlighting critical virus-receptor pairs.

Methodology:

Integrates known virus-receptor interaction data with receptor amino acid sequences; computes initial interaction scores from neighboring interactions; calculates virus and receptor GIP similarities via a Gaussian Interaction Profile kernel; computes receptor sequence similarity and integrates similarities with priority to sequence similarity; refines predictions using a Laplacian regularized least squares algorithm; validates performance with 10-fold cross-validation and leave-one-out cross-validation reporting AUCs (10CV 0.8675, LOOCV 0.9061) and comparisons to BRWH, LapRLS, and CMF.